DeepSeek Search & Fetch API
ActiveDeepSeek's pay-per-request web search and page fetch endpoints for agents
DeepSeek's Search and Fetch API gives agents web search and page fetch for $0.75 per 1,000 successful requests each, with empty results and failures not billed. Conditional: the guide we read states no data-retention or privacy terms and no response-length limits, and calls through DeepSeek Harness, Claude Code or Codex stop being free at 00:00 Beijing time on November 1, 2026. It suits teams already on the DeepSeek API that want metered search without another vendor. We have not tested result quality.
DeepSeek's October 17, 2026 update calls its in-house engine AI-native, and Search returns a rerank_score combining relevance, authority and freshness plus a query-related summary, but the guide we read does not say how those are produced.
Is it right for you?
Good for
- Giving an agent live web search and page reading from the same vendor as its model: both endpoints sit on the DeepSeek API and use the same bearer key
- Predictable billing: $0.75 per 1,000 successful requests for each endpoint, not by tokens, with empty results and failures not billed
- Controlled queries: Search supports site:, -site:, lang:, after: and before: operators and a count of 1 to 50, and can return page content with need_content
- Turning pages into Markdown for a model: Fetch returns the title and content as Markdown for any HTTP or HTTPS URL up to 8,192 characters
Not good for
- Teams that need published privacy or retention terms: the guide we read states none
- High-throughput crawling: each endpoint is capped at 5 queries per second per account, and Fetch also has a per-domain limit
- Assuming the Harness, Claude Code and Codex calls stay free: DeepSeek says per-request billing starts at 00:00 Beijing time on November 1, 2026
- Buyers who need independent quality evidence: the index size and spam-filtering claims are DeepSeek's own, and the pages we read include no third-party evaluation
Our experience
Pricing
Paid- Billed per successful request, not by tokens or results returned
- Empty results and failures not billed
- 5 QPS per account
- Returns page content as Markdown
- 5 QPS per account plus a per-domain limit
- HTTP 429 above a limit
- Server-side tool calls on the DeepSeek API through DeepSeek Harness, Claude Code or Codex
- Per-request billing starts at 00:00 Beijing time on November 1, 2026
Workflow recipes
Search, then fetch a result
Two documented calls: find pages, then read one as Markdown
- Send POST https://api.deepseek.com/v1/tools/search with an Authorization: Bearer header and a JSON body containing query (1 to 128 characters), optionally count and need_content
- Pick a result URL from the response (title, url, snippet, summary, authority and date are returned)
- Send POST https://api.deepseek.com/v1/tools/fetch with that URL (up to 8,192 characters) to get the title and Markdown content
- Wait and retry on HTTP 408, 429 or 5xx; do not retry 400, 401 or 413 without changing the request; handle 402 as insufficient balance and 403 as a security rejection
No verdict changes yet
The clock starts day one. Changes land here as our verdict evolves.
Sources
Verification log
No verification checks yet
We haven't logged a verification check for this entry. Once a check runs, its history shows here.
Want this running in your business?
We tested this tool and we build with tools like it every day. Tell us the workflow and we will set it up, integrate it and hand it over working.
The conversation runs on Gnosari, one of the tools in this directory. A real conversation, not a sales script.